DALL-E 2 is an ambitious project by Open AI which aims to convert natural language input into realistic images using AI. Marques Brownlee is one of the few people who tried it to check its claims. To his surprise, the system was able to generate photos that resembled the input text.
DALL-E 2 is not available to the general public and is a closed project backed by Elon Musk. Marques Brownlee was able to test it out by inputting multiple complex image requests and even made his video thumbnail using the system.
What kind of images did DALL-E 2 produce?
Marques began with a less complex image request of a blue apple in a bowl of oranges. The AI system generated ten images of a blue apple in a bowl of oranges with realistic textures. Next, he tried to generate images of an elderly kangaroo. The system produced ten convincing images based on that input text.
He even entered a request for a wise elephant staring at the moon at night. DALL-E 2 didn’t disappoint and produced ten images of an elephant staring at the moon.
Marques then upped the complexity a little bit with his next request. He asked the system to generate an image of a teddy bear doing surgery on grapes in a ’90s cartoon theme. DALL-E 2 produced images in an exact style that looked convincing.
However, when he tried to look into the details, he found some inherent flaws.
DALL-E 2 can produce images based on natural language input that looks photorealistic. But it misses some details like the relative position of the object in an image. When Marques asked it to produce images of a blue apple, the system didn’t keep Apple on top in every image.
So, it struggles with the relative position of an image but that may be fixed with DALL-E 3( if it happens). DALL-E 2 also struggles with text in images and fails to produce meaningful words in images. The AI system fails to generate accurate signboards with meaningful text.
Will it destroy the careers of graphic designers?
Marques addressed the question in his video and said that DALL-E 2 isn’t a commercial tool. However, it could be helpful in creating visualizations of a concept and then letting the graphic designer move forward with a vision in mind.
Last month, a person generated a tattoo with the help of DALL-E 2. Image production is just the beginning. The AI system could be capable of generating short clips, videos, and even full-length movies in the future. What do you think of DALL-E 2? Will it help graphic designers? Share your thoughts in the comments.